load_dataset doesn't support list column

[arthasking123]

Describe the bug

dataset = load_dataset("Doraemon-AI/text-to-neo4j-cypher-chinese")

got exception: Generating train split: 1834 examples [00:00, 5227.98 examples/s] Traceback (most recent call last): File "/usr/local/lib/python3.11/dist-packages/datasets/builder.py", line 2011, in _prepare_split_single writer.write_table(table) File "/usr/local/lib/python3.11/dist-packages/datasets/arrow_writer.py", line 585, in write_table pa_table = table_cast(pa_table, self._schema) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.11/dist-packages/datasets/table.py", line 2295, in table_cast return cast_table_to_schema(table, schema) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.11/dist-packages/datasets/table.py", line 2254, in cast_table_to_schema arrays = [cast_array_to_feature(table[name], feature) for name, feature in features.items()] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.11/dist-packages/datasets/table.py", line 2254, in arrays = [cast_array_to_feature(table[name], feature) for name, feature in features.items()] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.11/dist-packages/datasets/table.py", line 1802, in wrapper return pa.chunked_array([func(chunk, *args, **kwargs) for chunk in array.chunks]) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.11/dist-packages/datasets/table.py", line 1802, in return pa.chunked_array([func(chunk, *args, **kwargs) for chunk in array.chunks]) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.11/dist-packages/datasets/table.py", line 2018, in cast_array_to_feature casted_array_values = _c(array.values, feature[0]) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.11/dist-packages/datasets/table.py", line 1804, in wrapper return func(array, *args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.11/dist-packages/datasets/table.py", line 2115, in cast_array_to_feature raise TypeError(f"Couldn't cast array of type\n{array.type}\nto\n{feature}") TypeError: Couldn't cast array of type struct<m.name: string, x.name: string, p.name: string, n.name: string, h.name: string, name: string, c: int64, collect(r.name): list<item: string>, q.name: string, rel.name: string, count(p): int64, 1: int64, p.location: string, max(n.name): null, mn.name: string, p.time: int64, min(q.name): string> to {'q.name': Value(dtype='string', id=None), 'mn.name': Value(dtype='string', id=None), 'x.name': Value(dtype='string', id=None), 'p.name': Value(dtype='string', id=None), 'n.name': Value(dtype='string', id=None), 'name': Value(dtype='string', id=None), 'm.name': Value(dtype='string', id=None), 'h.name': Value(dtype='string', id=None), 'count(p)': Value(dtype='int64', id=None), 'rel.name': Value(dtype='string', id=None), 'c': Value(dtype='int64', id=None), 'collect(r.name)': Sequence(feature=Value(dtype='string', id=None), length=-1, id=None), '1': Value(dtype='int64', id=None), 'p.location': Value(dtype='string', id=None), 'substring(h.name,0,5)': Value(dtype='string', id=None), 'p.time': Value(dtype='int64', id=None)}

The above exception was the direct cause of the following exception:

Traceback (most recent call last): File "/home/ubuntu/llm/train-2.py", line 150, in dataset = load_dataset("Doraemon-AI/text-to-neo4j-cypher-chinese") ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.11/dist-packages/datasets/load.py", line 2609, in load_dataset builder_instance.download_and_prepare( File "/usr/local/lib/python3.11/dist-packages/datasets/builder.py", line 1027, in download_and_prepare self._download_and_prepare( File "/usr/local/lib/python3.11/dist-packages/datasets/builder.py", line 1122, in _download_and_prepare self._prepare_split(split_generator, **prepare_split_kwargs) File "/usr/local/lib/python3.11/dist-packages/datasets/builder.py", line 1882, in _prepare_split for job_id, done, content in self._prepare_split_single( File "/usr/local/lib/python3.11/dist-packages/datasets/builder.py", line 2038, in _prepare_split_single raise DatasetGenerationError("An error occurred while generating the dataset") from e datasets.exceptions.DatasetGenerationError: An error occurred while generating the dataset

Steps to reproduce the bug

dataset = load_dataset("Doraemon-AI/text-to-neo4j-cypher-chinese")

Expected behavior

no exception

Environment info

python 3.11 datasets 2.19.0

[yucc-leon]

I encountered this same issue when loading a customized dataset for ORPO training, in which there were three columns and two of them were lists. I debugged and found that it might be caused by the type-infer mechanism and because in some chunks one of the columns is always an empty list ([]), it was regarded as list<item: null>, however in some other chunk it was list<item: string>. This triggered a TypeError running the function table_cast().

I temporarily fixed this by re-dumping the file into a regular JSON format instead of lines of JSON dict. I didn't dig deeper for the lack of knowledge and programming ability but I do hope some developer of this repo will find and fix it.